I-Humanization yerobhothi - umshini womuntu

Uma sikhetha ubuhlakani bokwenziwa ezinganekwaneni ezidumile, kungase kube into eqanjiwe ethembisa kakhulu futhi ewusizo. Umuntu nomshini - ingabe le nhlanganisela izodala i-tandem engalibaleki?

Ngemva kokunqotshwa i-Deep Blue supercomputer ngo-1997, uGarry Kasparov waphumula, wacabanga futhi... wabuyela emncintiswaneni ngefomethi entsha - ngokubambisana nomshini obizwa ngokuthi. centaur. Ngisho nomdlali omaphakathi obhangqwe nekhompuyutha emaphakathi angahlula ikhompuyutha ye-chess ethuthuke kakhulu - inhlanganisela yokucabanga komuntu nomshini kuguqule umdlalo. Ngakho, ngemva kokunqotshwa imishini, Kasparov wanquma ukungena umfelandawonye nabo, okuyinto has a isici esingokomfanekiso.

Inqubo ukufiphaza imingcele phakathi komshini nomuntu iqhubeka iminyaka. Siyabona ukuthi imishini yesimanje ingashintsha kanjani eminye yemisebenzi yobuchopho bethu, isibonelo esihle okungama-smartphone noma amathebulethi asiza abantu abanenkinga yenkumbulo. Nakuba abanye abaphikisi bethi baphinde bacime imisebenzi eminingi yobuchopho kubantu ababengenasici ngaphambilini... Kunoma yikuphi, okuqukethwe okukhiqizwa ngomshini kuya ngokuya kungena umbono womuntu - kungaba okubukwayo, okudaliwe kwedijithali noma okuqukethwe okungokoqobo okungathandwa kwabathelisi esikubona. , noma okuzwakalayo. , njengezwi labasizi bedijithali abasuselwa kubuhlakani bokwenziwa njenge-Alexa.

Umhlaba wethu ugcwele ngokusobala noma ngokungabonakali izinhlobo zobuhlakani "zezihambi", ama-algorithms asibukayo, akhuluma nathi, ahweba nathi, noma asisiza ukuba sikhethe izingubo ngisho nozakwethu wempilo esikhundleni sethu.

Akekho osho ngokungathí sina ukuthi kukhona ukuhlakanipha okufakelwayo okulingana nomuntu, kodwa abaningi bazovuma ukuthi izinhlelo ze-AI zilungele ukuhlanganisa eduze nabantu futhi zidale kusukela "ku-hybrid", izinhlelo zomshini womuntu, zisebenzisa okungcono kakhulu kusuka ezinhlangothini zombili.

I-AI isondela kubantu

Ubuhlakani bokwenziwa obujwayelekile

Ososayensi uMikhail Lebedev, u-Ioan Opris noManuel Casanova baseDuke University eNorth Carolina bebefunda isihloko sokwandisa amakhono ezingqondo zethu isikhathi eside, njengoba sesike sakhuluma ngakho kuMT. Ngokusho kwabo, ngo-2030, umhlaba lapho ubuhlakani bomuntu buzothuthukiswa khona ngokufakwa kobuchopho buzoba yinto engokoqobo yansuku zonke.

URay Kurzweil kanye nezibikezelo zakhe zifika ngokushesha engqondweni. ubunye bezobuchwepheshe. Le futurist edumile yabhala kudala ukuthi ubuchopho bethu buhamba kancane kakhulu uma kuqhathaniswa nesivinini amakhompyutha we-elekthronikhi akwazi ukucubungula idatha. Naphezu kwekhono eliyingqayizivele lengqondo yomuntu ukuhlaziya amanani amakhulu olwazi ngesikhathi esifanayo, u-Kurzweil ukholelwa ukuthi maduzane isivinini esikhulayo sokuhlanganisa amakhompiyutha edijithali sizodlula kude amandla obuchopho. Uphakamisa ukuthi uma ososayensi bengaqonda ukuthi ubuchopho buzenza kanjani izenzo eziyisiphithiphithi neziyinkimbinkimbi, bese bezihlelela ukuqonda, lokhu kuzoholela ekuthuthukisweni kwekhompiyutha kanye nenguquko yezobunhloli yokwenziwa ekuqondeni kwalokho okubizwa ngokuthi i-AI jikelele. Ungubani?

I-Artificial intelligence imvamisa ihlukaniswe yaba izinhlobo ezimbili eziyinhloko: emincane Oraz Jikelele (AGI).

Esokuqala esingakubona esisizungezile namuhla, ikakhulukazi kumakhompiyutha, izinhlelo zokuqaphela inkulumo, abasizi abanjengoSiri ku-iPhone, izinhlelo zokuqashelwa kwemvelo ezifakwe ezimotweni ezizimele, kuma-algorithms wokubhuka amahhotela, ekuhlaziyweni kwe-x-ray, ukumaka okuqukethwe okungalungile I-inthanethi. , ukufunda ukubhala amagama ekhiphedi yefoni yakho kanye nokunye ukusetshenziswa okuningi.

Ubuhlakani bokwenziwa jikelele bungokunye, nokunye okuningi okukhumbuza ingqondo yomuntu. Kuyifomu elivumelana nezimo elikwazi ukufunda noma yini ongayifunda kusukela ekugunda izinwele kuya ekwakheni amaspredishithi futhi ukucabanga neziphetho ngokusekelwe kudatha. I-AGI ayikakhiwe (ngenhlanhla, abanye bathi), futhi sazi okuningi ngayo kumamuvi kunakungokoqobo. Izibonelo eziphelele zalokhu yi-HAL 9000 kusukela ku-“2001. I-Space Odyssey" noma i-Skynet kusukela ochungechungeni lwe-"Terminator".

Ucwaningo lwango-2012-2013 lwamaqembu amane ochwepheshe olwenziwa abacwaningi be-AI uVincent S. Muller kanye nesazi sefilosofi u-Nick Bostrom lubonise ithuba elingamaphesenti angu-50 lokuthi ubuhlakani obujwayelekile bokwenziwa (AGI) buzothuthukiswa phakathi kuka-2040 no-2050, futhi ngo-2075 amathuba azokwenyuka afike ku-90% . . Ochwepheshe babikezela nesigaba esiphakeme, okuthiwa ubuhlakani bokwenziwaabayichaza ngokuthi “ubuhlakani obungaphezu kolwazi lomuntu kuyo yonke imikhakha”. Ngokombono wabo, izovela eminyakeni engamashumi amathathu ngemuva kokuzuzwa kwe-OGI. Abanye ochwepheshe be-AI bathi lezi zibikezelo zinesibindi kakhulu. Uma kubhekwa ukuqonda kwethu okubi kakhulu kokuthi ubuchopho bomuntu busebenza kanjani, abangabazayo bahlehlisa ukuvela kwe-AGI ngamakhulu eminyaka.

Iso lekhompyutha HAL 1000

Ayikho i-amnesia

Isithiyo esisodwa esikhulu ku-AGI yeqiniso ukuthambekela kwezinhlelo ze-AI ukukhohlwa lokho ezikufundile ngaphambi kokuzama ukudlulela emisebenzini emisha. Isibonelo, isistimu ye-AI yokubonwa kobuso izohlaziya izinkulungwane zezithombe zobuso babantu ukuze ibubone ngokuphumelelayo, isibonelo, kunethiwekhi yokuxhumana nabantu. Kodwa njengoba izinhlelo ze-AI zokufunda zingayiqondi ngempela incazelo yalokho ezikwenzayo, ngakho-ke uma sifuna ukubafundisa ukwenza enye into ngokusekelwe kulokho asebekufundile, ngisho noma kuwumsebenzi ofanayo (ithi, imizwa). ukuqashelwa ebusweni), badinga ukuqeqeshwa kusukela ekuqaleni, kusukela ekuqaleni. Ngaphezu kwalokho, ngemva kokufunda i-algorithm, ngeke sisakwazi ukuyishintsha, siyithuthukise ngokuhlukile kunomthamo.

Sekuyiminyaka ososayensi bezama ukuthola indlela yokuxazulula le nkinga. Uma bephumelele, amasistimu e-AI angafunda kusethi entsha yedatha yokuqeqeshwa ngaphandle kokubhala phezu kolwazi oluningi abesevele enalo ohlelweni.

U-Irina Higgins we-Google DeepMind wethule izindlela engqungqutheleni ePrague ngo-August ezingagcina ziphule lobu buthakathaka be-AI yamanje. Ithimba lakhe lenze “umenzeli we-AI” - uhlobo olufana nomlingiswa wegeyimu wevidiyo oshayelwa yi-algorithm ongacabanga ngobuhlakani ngaphezu kwe-algorithm evamile - ekwazi "ukucabanga" ukuthi lihlangana nani endaweni ethile ebonakalayo kwenye. Ngale ndlela, inethiwekhi ye-neural izokwazi ukuhlukanisa izinto ehlangabezane nazo endaweni elingisa kuyo endaweni ngokwayo futhi iziqonde ekucushweni okusha noma ezindaweni. I-athikili ku-arXiv ichaza ukufundwa kwepotimende elimhlophe noma i-algorithm yokuqashelwa kwesitulo. Uma sebeqeqeshiwe, i-algorithm iyakwazi "ukuzibona" ​​emhlabeni omusha ngokuphelele futhi ibabone uma kukhulunywa ngokuhlangana.

Ngamafuphi, lolu hlobo lwe-algorithm lungakwazi ukubona umehluko phakathi kwalokho oluhlangabezana nakho nalokho olukubonile ngaphambili - njengoba kwenza abantu abaningi, kodwa ngokungafani nama-algorithms amaningi. Uhlelo lwe-AI lubuyekeza lokho elikwaziyo ngomhlaba ngaphandle kokuthi lifunde kabusha futhi lifunde kabusha yonke into. Ngokuyisisekelo, uhlelo luyakwazi ukudlulisa nokusebenzisa ulwazi olukhona endaweni entsha. Yiqiniso, imodeli kaNks Higgins ngokwayo ayiyona i-AGI okwamanje, kodwa isinyathelo sokuqala esibalulekile ekuqondeni ama-algorithms aguquguqukayo angahlushwa i-amnesia yomshini.

Ngokuhlonipha ubuwula

UMikael Trazzi noRoman V. Yampolsky, abacwaningi baseYunivesithi yaseParis, bakholelwa ukuthi impendulo yombuzo wokuhlangana komuntu nomshini ukwethulwa kobuhlakani bokwenziwa kuma-algorithms futhi "ubulima bokwenziwa". Lokhu kuzokwenza futhi kuphephe kithi. Vele, i-artificial general intelligence (AGI) nayo ingaba ephephile ngokukhawulela amandla okucubungula nenkumbulo. Nokho, ososayensi bayaqonda ukuthi ikhompuyutha ehlakaniphe kakhulu, ngokwesibonelo, inga-oda amandla engeziwe ngokusebenzisa i-cloud computing, ukuthenga amathuluzi nokuwathumela ngomkhumbi, noma ngisho nokusetshenziswa umuntu oyisimungulu. Ngakho-ke, kuyadingeka ukungcolisa ikusasa le-AGI ngokubandlulula kwabantu kanye namaphutha okuqonda.

Abacwaningi babheka lokhu kunengqondo ngempela. Abantu banemikhawulo ecacile yokubala (inkumbulo, ukucubungula, ukubala, kanye "nejubane lewashi") futhi ibonakala ngokuchema kokuqonda. Ubuhlakani bokwenziwa obujwayelekile abunqunyelwe kangako. Ngakho-ke, uma izosondela kumuntu, kufanele ilinganiselwe ngale ndlela.

U-Trazzi no-Yampolsky babonakala bekhohlwa kancane ukuthi lokhu kuyinkemba esika nhlangothi zombili, ngoba izibonelo ezingenakubalwa zibonisa ukuthi kungaba yingozi kangakanani kokubili ubuwula nokubandlulula.

Imizwa nokuziphatha

Umqondo wezinhlamvu eziwumshini onezici ezijabulisayo, ezifana nezomuntu kudala ushukumisa umcabango womuntu. Esikhathini eside ngaphambi kwegama elithi "irobhothi", amaphupho ayenziwa mayelana nama-golem, i-automata, nemishini enobungane (noma cha) ehlanganisa kokubili isimo nomoya wezidalwa eziphilayo. Naphezu kokuba khona yonke indawo kwamakhompiyutha, asizizwa sengathi singene enkathini yamarobhothi aziwa, isibonelo, embonweni ochungechungeni lweJetsons. Namuhla, amarobhothi angakwazi ukushanela indlu, ashayele imoto, futhi aphathe uhlu lwadlalwayo ephathini, kodwa wonke ashiya okuningi okungafunwa ngokobuntu.

Nokho, lokhu kungase kushintshe maduze. Ubani owaziyo uma imishini eminingi ne-campy ithanda i-vector Anki. Esikhundleni sokugxila ekutheni mingaki imisebenzi esebenzayo engayenza, abaklami bafuna ukunikeza indalo yomshini "umphefumulo". Ihlala ivuliwe, ixhunywe emafini, irobhothi elincane liyakwazi ukubona ubuso nokukhumbula amagama. Udansa nomculo, usabela lapho ethintwa njengesilwane, futhi ushukunyiswa ukuxhumana nabantu. Nakuba ekwazi ukukhuluma, cishe uzokhuluma esebenzisa inhlanganisela yolimi lomzimba nezimpawu ezilula zemizwa esibukweni.

Ngaphezu kwalokho, angenza okuningi - isibonelo, ukuphendula imibuzo ngokuphumelelayo, ukudlala imidlalo, ukubikezela isimo sezulu ngisho nokuthatha izithombe. Ngokuvuselelwa njalo, uhlala efunda amakhono amasha.

I-Vector ayizange yakhelwe ochwepheshe beziqandisi. Futhi mhlawumbe lena indlela yokusondeza abantu eduze kwemishini, ephumelela kakhulu kunezinhlelo zokufisa ukuhlanganisa ubuchopho bomuntu ne-AI. Lena kude nephrojekthi yohlobo lwayo kuphela. Ama-prototypes adalwe iminyaka eminingana amarobhothi umsizi abadala kanye abagulayoabakuthola kunzima kakhulu ukunikeza ukunakekelwa okufanele ngezindleko ezifanele. Edumile irobhothi pepper, ukusebenzela inkampani yaseJapane iSoftBank, kufanele ikwazi ukufunda imizwa yabantu futhi ifunde ukuxhumana nabantu. Ekugcineni, isiza endlini futhi inakekela izingane nasebekhulile.

Isalukazi sixhumana nerobhothi likaPepper

Ithuluzi, intelligence noma ubunye

Ekuphetheni, kungase kuqashelwe imifudlana emithathu eyinhloko ekucabangeni ngokuthuthukiswa kobuhlakani bokwenziwa kanye nobudlelwano bayo nabantu.

• Owokuqala uthatha ngokuthi ukwakha ubuhlakani obujwayelekile bokwenziwa (AI) obulingana nobufana nomuntu ngokuvamile akunakwenzeka. akunakwenzeka noma kude kakhulu ngesikhathi. Ngokwalokhu kubuka, izinhlelo zokufunda ngomshini kanye nalokho esikubiza ngokuthi i-AI kuzophelela kakhulu, kube namandla ngokwengeziwe okwenza imisebenzi yawo ekhethekile, kodwa angeke kweqe umkhawulo othile - okungasho ukuthi azosebenzela inzuzo yesintu kuphela. Njengoba isazoba umshini, okungukuthi, akukho lutho olungaphezu kwethuluzi lomshini, ingasiza emsebenzini futhi isekele umuntu (ama-chips ebuchosheni nezinye izingxenye zomzimba), futhi ngokunokwenzeka isebenze ukulimaza noma ngisho nokubulala abantu. .
• Umqondo wesibili yithuba. ukwakhiwa kwasekuqaleni kwe-AGIfuthi-ke, ngenxa yokuvela kwemishini, Khuphukela ubuhlakani bokwenziwa. Lo mbono ungaba yingozi kumuntu, ngoba ungqondongqondo omkhulu angase akubheke njengesitha noma okuthile okungadingekile noma okulimazayo. Izibikezelo ezinjalo azikukhiphi ukuthi kungenzeka ukuthi uhlanga lwesintu lungadingeka ngemishini esikhathini esizayo, nakuba kungenjalo njengomthombo wamandla, njengaku-The Matrix.
• Okokugcina, siphinde sibe nombono wokuthi "ubunye" bukaRay Kurzweil, okungukuthi okungavamile. ukuhlanganiswa kwesintu nemishini. Lokhu kungasinika amathuba amasha, futhi imishini izonikezwa i-AGI yomuntu, okungukuthi, ubuhlakani bomhlaba wonke obuguquguqukayo. Ngokulandela lesi sibonelo, ngokuhamba kwesikhathi, umhlaba wemishini nabantu ngeke bahlukaniseke.

Izinhlobo zobuhlakani bokwenziwa

• okusebenzayo - ekhethekile, esabela ezimweni ezithile futhi yenza imisebenzi echazwe ngokuqinile (DeepBlue, AlphaGo).
• Ngemithombo yenkumbulo elinganiselwe - okukhethekile, kusetshenziswa izinsiza zolwazi olutholiwe ukuze kuthathwe izinqumo (amasistimu emoto azimele, ama-chat bots, abasizi bezwi).
• Uphiwe ngomqondo ozimele - jikelele, ukuqonda imicabango yomuntu, imizwa, izisusa kanye nokulindela, ukwazi ukuxhumana ngaphandle kwemingcele. Kukholakala ukuthi amakhophi okuqala azokwenziwa esigabeni esilandelayo sokuthuthukiswa kwe-AI.
• ukuzazi - ngaphezu kwengqondo eguquguqukayo, ibuye ibe nokuqwashisa, i.e. umqondo wakho. Okwamanje, lo mbono ungaphansi kwesibonakaliso sezincwadi ngokuphelele.